Study guide
Planning a Strong Health Economics Assignment
A practical guide to structuring an assignment, presenting epidemiological calculations and interpreting observational evidence without overstating causality.
Study context
Independent learning material
This resource was prepared for study and revision. Any named institution describes the programme context; it does not imply that the institution published, endorsed or maintains this page.
A strong assignment makes its reasoning inspectable. The marker should be able to see how the question led to the structure, where the evidence came from, how each calculation was performed and why the interpretation does—or does not—follow from the study design.
This independent postgraduate study guide uses a simplified Basic Epidemiology exercise as a worked example because epidemiological measures support later work in health economics, economic evaluation and health-policy analysis. It is not an official University resource, model answer or submission-ready document.
Start with the assessment task
Read the question alongside the learning outcomes, marking criteria, word limit and required reference style. If any of these are missing, identify the outputs demanded by the question and create a structure that addresses each one. Do not add a long general introduction at the expense of assessed calculations or interpretation.
The illustrative task describes a 20-year cohort study of 2,000 people. Among 800 alcohol users, 160 developed oesophageal cancer. Among 1,200 non-users, 40 developed the outcome. The task asks for a 2×2 table, incidence in each group, relative risk, attributable risk, attributable risk percent and a discussion of confounding.
The figures are hypothetical. They show an association in an observational design; they do not establish that alcohol caused each recorded case.
Find and use an authoritative definition
Begin with a source that matches the method. The CDC’s epidemiology guidance explains that the risk ratio is commonly used for cohort data because risks can be calculated for exposed and unexposed groups. The original study note also drew on Alexander, Lopes, Ricchetti-Masterson and Yeatts’ ERIC Notebook (2nd ed., 2014), published by the UNC Gillings School of Global Public Health.
Paraphrasing means reconstructing an idea accurately in your own analytical context, not replacing a few words with synonyms. Cite the source even when the sentence is fully paraphrased.
Parenthetical citations
In a parenthetical citation, the author and year sit within parentheses after the supported proposition. For example:
Cohort designs permit direct comparison of risk between groups defined by exposure status (Alexander et al., 2014).
The citation should follow the claim it supports. A citation at the end of a long paragraph may leave the reader uncertain about which sentences came from that source.
Narrative citations
In a narrative citation, the author forms part of the sentence and the year normally follows the author name:
Alexander et al. (2014) describe cohort studies as designs that follow groups over time to observe subsequent outcomes.
APA Style distinguishes these two citation forms. Choose between them according to the sentence’s emphasis. Neither form removes the need to represent the source accurately.
Construct the 2×2 table
Calculate the non-cases before attempting the effect measures:
- exposed non-cases:
800 − 160 = 640; - unexposed non-cases:
1,200 − 40 = 1,160.
| Alcohol-use status | Oesophageal cancer | No oesophageal cancer | Total |
|---|---|---|---|
| Users (exposed) | 160 | 640 | 800 |
| Non-users (unexposed) | 40 | 1,160 | 1,200 |
| Total | 200 | 1,800 | 2,000 |
Give the table a descriptive title in an assignment and state that the figures come from the question. A table should allow the reader to reproduce every subsequent calculation.
Show and interpret each calculation
The CDC defines incidence proportion as new cases during a specified period divided by the population at risk at the start of that period.
1. Risk in each group
Risk among exposed = 160 ÷ 800 = 0.20 = 20%
Risk among unexposed = 40 ÷ 1,200 = 0.03333 = 3.33%
Over the stated follow-up, 20% of the exposed group and 3.33% of the unexposed group developed the outcome in this hypothetical dataset. Calling these quantities incidence rates would be imprecise because the task provides people at risk, not person-time.
2. Risk ratio
RR = 0.20 ÷ 0.03333 = 6.0
The exposed group had six times the observed risk of the outcome compared with the unexposed group. This is a measure of association. Without information about selection, measurement, loss to follow-up, confounding and statistical uncertainty, it is not a causal estimate.
3. Attributable risk, or risk difference
Risk difference = 0.20 − 0.03333 = 0.16667
The observed excess risk was 16.67 percentage points, equivalent to approximately 167 additional cases per 1,000 exposed people over the stated follow-up. This absolute measure should not be described as a 16.67% relative increase.
4. Attributable risk percent among the exposed
ARP = (0.20 − 0.03333) ÷ 0.20 × 100 = 83.33%
The corrected result is 83.33%. The CDC calls this the attributable proportion. Its causal interpretation requires the observed association to be causal and adequately controlled for other causes. Under those assumptions, 83.33% of the risk among exposed people could be attributed to the exposure. Without those assumptions, it is safer to report the arithmetic and label it an attributable proportion calculated from the observed risks.
Discuss confounding rather than listing it
A confounder is associated with the exposure, independently affects the outcome and is not an intermediate step on the causal pathway under study. Smoking is an obvious candidate because it may differ between alcohol users and non-users and is independently associated with oesophageal cancer. Age, diet and socioeconomic position may also matter, but each proposed variable should be justified rather than copied into a generic list.
Explain the consequence: if smoking is more common among alcohol users, an unadjusted risk ratio may partly reflect smoking rather than the independent association with alcohol. Restriction and matching can address selected confounders at the design stage; stratification and multivariable modelling can address measured confounders during analysis. None of these methods automatically removes residual confounding or error in the confounder measurements.
Build a proportionate structure
For a short calculation-based assignment, a workable structure is:
- a brief introduction defining the design and the analytical task;
- the 2×2 table and assumptions;
- one subsection for each effect measure, containing the definition, formula, calculation and interpretation;
- a focused discussion of confounding and other limitations; and
- a short conclusion answering what the data support.
Include a contents page, list of tables or abbreviations only when the assignment’s length and instructions justify them. A three-page exercise rarely needs the front matter expected in a dissertation. Clear headings, numbered tables and consistent notation matter more than decorative complexity.
Reference the evidence actually used
Follow the edition specified by the module. APA 6 and APA 7 differ in reference-list presentation, so do not combine them. The APA Style author–date guidance is the controlling public source for current APA 7 in-text conventions.
For this study exercise, the core references are:
- Alexander, L. K., Lopes, B., Ricchetti-Masterson, K., & Yeatts, K. B. (2014). Cohort studies. ERIC Notebook Series (2nd ed.). UNC Gillings School of Global Public Health.
- Centers for Disease Control and Prevention. (n.d.). Analyzing and interpreting data. In The CDC Field Epidemiology Manual. Original
Check every reference against the source itself. Do not cite a paper merely because another author cited it, and do not include sources that never informed the assignment.
Academic integrity and AI-assistance disclosure
The original Research Mind page stated that parts of its material had been generated using ChatGPT 5.2 Pro and Gemini 3. This migrated resource retains that disclosure. During migration, the example was restructured, the calculations were recomputed, the attributable risk percent was corrected to 83.33%, and causal language was qualified against CDC guidance.
AI-generated explanations and calculations can be wrong. Learners should verify them against the cited sources and disclose any use required by their institution. Nothing on this page should be copied and submitted as original assessed work. The source’s downloadable PDF and Word answer files are deliberately not republished.